Digital health interventions, such as remote patient monitoring, have the potential to revolutionize patient care by providing timely and personalized interventions that improve patient outcomes while optimizing resource allocation and reducing healthcare costs. However, the development, evaluation, and optimization of these interventions remain challenging due to limited data, high-dimensional state and action spaces, and the need for interpretable, clinically-grounded treatment policies. This dissertation addresses these challenges through the development, evaluation, and optimization of a remote patient monitoring platform for youth with type 1 diabetes in the 4T Study at Stanford Children's Hospital. First, it describes how the platform was developed to reduce provider burden by incorporating algorithm-enabled patient prioritization based on consensus glucose targets to identify the patients most in need of provider attention. The introduction of new prioritization methods and an interactive dashboard substantially reduced provider review time per patient and increased weekly clinic capacity. Patients receiving remote monitoring demonstrated improved glucose control compared to those without remote monitoring. Second, the dissertation introduces Smart Start, a novel approach to designing more powerful clinical trials of digital health interventions by optimizing the intervention with pilot data before evaluating it in a randomized trial. By leveraging natural variation in patient eligibility for remote monitoring during the 4T pilot study, Smart Start identifies an optimized intervention with twice the estimated effect size of an unoptimized intervention. This approach enables the design of a powerful clinical trial with far fewer participants, maintaining a high probability of success while substantially reducing the trial's cost. Finally, the dissertation presents π-STAR, a practical approach to policy learning for digital health interventions that combines clinician expertise with causal machine learning to develop interpretable and effective treatment policies. In the 4T study's remote monitoring setting, policies derived from clinician-informed representations significantly outperform those learned from raw data in terms of interpretability and efficacy. These findings suggest that many digital health interventions could be substantially improved by learning interpretable policies from clinician-informed representations of patient states and interventions. Together, these contributions highlight the importance of integrating clinical knowledge and feedback, algorithm-enabled prioritization, and clinician-informed policy learning in the development of digital health interventions. The approaches presented contribute to the field by offering scalable, interpretable, and effective solutions for remote patient monitoring, clinical trial design, and policy learning, ultimately enabling solutions that can improve patient outcomes and optimize resource utilization. Future research should focus on scaling these interventions, evaluating them in randomized trials, applying them to other chronic conditions, and ensuring equitable access to optimized interventions.